AI product Open source · MIT
Hex is a local-first voice-dictation application developed by Anomaly for Apple silicon Macs running macOS 15 or newer, with beta releases for Windows and Linux. Users hold or toggle a shortcut, speak, and release it; Hex transcribes the audio with a locally downloaded speech model and pastes the result into the focused application. It supports configurable shortcuts, languages, microphones, models, recording behavior, application-specific Modes, correction rules, optional text transformations, dictation history, and command-line transcript queries.
The macOS and Linux versions also include Annotate, which records the screen while the user draws and speaks, then produces a recording containing keyframes, drawings, and a transcript for pasting or sending to an OpenCode session. OpenCode integration is optional and can post-process transcripts or receive annotations through a separately running OpenCode service. Advanced users can define deterministic TypeScript or JavaScript transformations in ~/.config/hex/hex.config.ts; these scripts run in an embedded JavaScript engine without filesystem, network, or process access. The project also publishes a TypeScript SDK through npm for applications that provide recorded audio and use Hex for model preparation and transcription.
Speech transcription runs locally, and dictation audio is not saved by default. Optional OpenCode features send transcript and context data to the configured model provider, while model installation and software updates require network access. Hex is distributed for macOS, Windows, and Linux under the MIT License.